An Introduction to JVM Performance

on YouTube ↗01h 01m

" summary

HotSpot runs methods at levels 0–4 using C1/C2 JIT with runtime profiling and speculative inlining, and about 90% of call sites are monomorphic, so inlining and devirtualization drive most speedups. Avoid megamorphic call sites and excess anonymous types; object pooling usually hurts because escape analysis removes short‑lived objects and avoids heap allocation. Measure with harnesses like JMH or Mission Control and inspect intrinsics and machine code (adopt's "watch"); false sharing and cache misses (reading large lines costs millions of instructions) cause far bigger slowdowns than micro-optimizations.

AI-written summary · about

" chapters (13)

1. 0:00Start
2. 2:56How a Java Program Is Run
3. 4:50The Jvm Layer and the Processor Layer
4. 5:52How Does Hotspot Run Your Code
5. 14:35Single Inheritance
6. 20:23Mega Morphic Call Sites
7. 31:50Inheritance Model
8. 39:45Code Elimination
9. 45:37Intrinsics
10. 47:08Sorting Algorithm
11. 50:15False Sharing
12. 55:17Heisenberg's Uncertainty Principle
13. 56:46Safe Points

" related talks

• 41mAn Introduction to JVM PerformanceRafael Winterhalter · GOTO Oslo 2020
• 59mAn Introduction to JVM PerformanceRafael Winterhalter · JavaDay Lviv 2016
• 49mAn Introduction to JVM PerformanceRafael Winterhalter · Devoxx Belgium 2017
• 54mJava performance techniques The cost of HotSpot runtime optimizationsIonut Balosin · Voxxed Days Vienna 2017
• 35mJava at Speed: Building a Better JVMSimon Ritter · JCON 2020
• 01h 09mWhat it means to be Just In TimeAshley Puls · SpringOne2GX 2015
• 01h 09mFrom Bytecode to Native Code in HotSpot JVMIonut Balosin · Voxxed Days Bucharest 2016
• 11mJVM internals — bytecode, JIT, GC, speculative optimizationGerrit Grunwald · JCON EUROPE 2026